Cooperative task unloading method in mobile Internet of Vehicles

Abstract

The invention relates to a cooperative task unloading method in a mobile internet of vehicles, and belongs to the field of intelligent driving. The method is implemented by a mobile Internet of Vehicles system, and comprises the following steps: S1, a mobile device acquires a generated task and the length of a task waiting queue in real time; s2, the mobile device judges whether the task needs to be unloaded or not; s3, tasks which cannot be unloaded by the mobile equipment are directly unloaded to the cloud server; s4, determining whether the task is unloaded to the edge server or the cloud server by utilizing a DDQN-based mobile equipment side task unloading method; s5, determining an actual execution position of the task based on an A3C distributed method; and S6, executing and returning a result to the mobile equipment. According to the method, the misunderstanding of selecting the optimal computing server from the nearby edge server in the traditional method is broken through, the task unloading process is divided into a task unloading method deployed on the mobile device side and the edge server side, efficient task unloading and scheduling can be achieved, and optimal configuration of resources is achieved.

Description

Technical Field

[0001] The present invention relates to a method for unloading collaborative tasks in a mobile vehicle network, belongs to the field of intelligent driving, and in particular to unloading collaborative tasks in a mobile vehicle network. Background Art

[0002] In daily task computing scenarios, tasks generated by both vehicles and smart mobile terminals have completion time limits. Too high transmission delays or computing delays can lead to task computing failures and a significant drop in user service quality. However, relying solely on the limited computing power of the mobile terminal itself may not be able to complete these computing tasks. Therefore, the key issue is how to efficiently offload tasks and utilize the computing resources of edge computing servers. Thanks to the development of mobile edge computing technology, edge computing devices are placed on many roads and areas. Vehicles quickly offload tasks to edge computing servers through communication technology between vehicles and infrastructure, and quickly complete task computing with the help of the stronger computing power of edge devices.

[0003] Obviously, offloading tasks with large computing requirements to edge computing devices is the best choice, but some small tasks or tasks with high delay tolerance can also be completed by local computing. This leads to two urgent problems that need to be solved in the offloading process: whether the task should be offloaded to the cloud server or edge server for computing and which edge server the task should be offloaded to for computing. Traditional task offloading algorithms are divided into centralized and distributed types. Most current studies directly offload tasks based on the current situation, and the offloading location is any edge computing server in the edge server cluster, which is inconsistent with the actual scenario. Because the coverage range and signal transmission power of the edge server are limited, tasks cannot be offloaded to any edge server in the edge server cluster at will. In addition, issues such as cross-domain vehicles also restrict the efficiency of the task offloading algorithm.

[0004] Even if the task is offloaded to the edge computing server, if the server is highly loaded and the computing queue is severely blocked, the offloading result of the task will become counterproductive. The unknown load status of the edge server has a great interference with the offloading of tasks, and is even the main reason for the failure of task offloading. Secondly, if the task can be completed by calculation, but the device has moved out of the coverage of the edge server, how to transmit the result to the device is also a factor that limits task offloading in real scenarios. Solving these problems has become a hot topic in the research of task offloading algorithms. Summary of the invention

[0005] The purpose of the present invention is to propose a more efficient task offloading method compared with the prior art, and to provide a collaborative task offloading method in a mobile vehicle network, by dividing the traditional task offloading process into two stages: the task offloading decision stage and the task calculation redirection stage. Among them, the mobile device side algorithm is only responsible for making an offloading decision on whether to offload the task according to the current device status, and if offloading, whether to offload to the cloud or the edge server, and the edge computing server side algorithm makes a decision on the actual calculation location of the task according to the load status of the server cluster.

[0006] To achieve the above object, the present invention discloses a method for unloading collaborative tasks in a mobile vehicle network. Figure 1 , which is implemented by a mobile vehicle networking system, which includes: several mobile devices, several edge servers, and a cloud server; the mobile device is installed on the car terminal and connected to the edge server and the cloud server through a wireless network; the wireless network connecting the mobile device and the cloud server includes two parts: a 5G network and a core network, which is first connected to the base station through the 5G network and then the base station offloads the task to the cloud server connected through the core network; the edge server is connected to the cloud server through a wired network; the mobile device, edge server, and cloud server are all computers.

[0007] For M mobile devices V = {v i |1≤i≤M}, N edge servers S={s j |1≤j≤N}, a scenario with 1 cloud server c, for any mobile device v i The parameters are Any edge server j The parameters include The parameters of cloud server c include [f c , ρ c ]; where f * Indicates the computing power of the device, that is, the number of cycles that the CPU can execute in one second; c * Represents the storage capacity of the device; ρ * Represents the total number of signal channels of the device; * indicates the corresponding device, i.e., mobile device, edge server or cloud server; and Corresponding to the mobile device v at time t i Locations and Edge Servers j location; For mobile devicesv i The speed at time t; For edge servers j The wireless network coverage radius; For edge servers jThe load vector from time tn to time t, where n is a positive integer set according to the storage capacity of the device; set any task task generated by all mobile devices at time t k (t) = [e k (t), r k (t), d k (t),u k (t)], where k∈K, K is the set of tasks generated by all mobile devices, e k (t) represents the size of the task at time t, r k (t) represents the number of CPU cycles required to complete the task calculation at time t, u k (t) represents the size of the result returned after the task is calculated at time t, d k (t) represents the deadline of the task at time t, that is, at time t+d k If the result is not returned to the mobile device, the task will be considered as a calculation failure; the 5G network transmission speed is recorded as b cell , the transmission speed of the core network is recorded as b core .

[0008] The present invention also discloses a method for unloading collaborative tasks in a mobile vehicle network, combining Figure 2 , including the following steps:

[0009] S1: The mobile device obtains the generated tasks and the length of the task waiting queue in real time;

[0010] S2: The mobile device calculates the time required to complete the task based on its own computing power. If the task can be completed within the deadline, the task is processed locally. Otherwise, a task offloading decision needs to be made.

[0011] S3: If the mobile device is outside the coverage area of ​​the edge server, or the task cannot be offloaded before the mobile device leaves the coverage area of ​​the edge server, the task is directly offloaded to the cloud server;

[0012] S4: When task offloading decision is needed, the mobile device-side task offloading method based on DDQN is used to decide whether to offload the task to the edge server or the cloud server, combining the location information of the mobile device and the edge server and the time information of sending the task;

[0013] S5: When the task is offloaded to the edge server, the distributed method based on A3C (Asynchronous Advantage Actor-Critic, A3C for short) will determine the actual execution location of the task according to the overall load status of the server cluster;

[0014] S6: Execute the computing task and send the result back to the mobile device.

[0015] Furthermore, since the task may not be transmitted yet, incomplete data should be avoided when sampling in step S1.

[0016] Further, at time t, the mobile device v described in step S2 i Calculate its task based on its own computing power k (t) Time required for completion Among them, mobile devices v i Computing tasks k Waiting time (t) x i，j / c,k Represents a mobile device v i Uninstall task task k To edge servers j Or cloud server c; if the device decides to offload the task to an edge server or cloud server, then x i,j / c,k =1; on the contrary, when deciding the local computing task, x i,j / c,k = 0. If the mobile device has enough storage space and T c ≤d k On mobile devices v i The task is processed locally.

[0017] Further, step S3 is specifically as follows: the distance between the mobile device and all edge servers is calculated according to the Euclidean distance formula. If these distances are greater than the wireless network coverage radius of all edge servers, the task is directly unloaded to the cloud server; at the same time, if the wireless network coverage requirements are met, the location at the task deadline is calculated according to the speed of the mobile device and the task deadline. If the location is out of the wireless network coverage of all edge servers, the task is directly unloaded to the cloud server.

[0018] Further, the DDQN-based mobile device side task offloading method described in step S4 is implemented using the DDQN network, and for the mobile device vi at time t, specifically:

[0019] S401: Calculate the distance between the mobile device and v according to the Euclidean distance formula i Nearest Edge Servers j ;

[0020] S402: Calculate the maximum transmission rate b of a single channel using the Shannon formula constraint, and further calculate the edge server s j Transmission speed

[0021] S403: Acquire the status of task offloading on the mobile device side by combining the location information of the mobile device and the edge server and the task sending time information in; For edge servers j Computing tasks k The waiting time (t) can be calculated based on the edge server s j The sum of the computation time of the tasks in the queue is calculated;

[0022] S404: Set rewards

[0023] S405: Combine the status of task offloading on the mobile device side and rewards Using DDQN (DoubleQ-learning Networks, DDQN for short) to select actions

[0024] S406: Follow the action Perform task offloading.

[0025] Furthermore, in order to prevent the polarization of edge server utilization, the load status of all edge servers at the current moment needs to be taken into account, thereby ensuring the load balancing of the entire server cluster. Therefore, the A3C-based distributed method described in step S5 is implemented using the A3C network, specifically:

[0026] S501: Predict the load vector of all edge servers at the next moment based on the historical load of all edge servers

[0027] Preferably, any edge server s j The predicted load at time t+1 is calculated based on the load from time tn to time t: Among them, f θ (·) is a neural network with parameter θ;

[0028] S502: Obtain the status of the A3C network Where T is the time when the task is generated;

[0029] S503: Selecting an action using the Actor network of the A3C network in, are positive integers, corresponding to the codes of the edge server and the cloud server respectively;

[0030] S504: Calculate reward based on task completion time setting

[0031] Specifically, (1) when the selected action is the distance v from the mobile device i Nearest Edge Servers j hour,

[0032]

[0033] Among them, φ k Indicates whether the mobile device is within the coverage of the current server after completing the task calculation. If yes, it is 0, otherwise it is 1; Indicates that the task is offloaded to the edge server s j The time waiting for calculation in the calculation queue is the measured value after the action is run;

[0034] (2) When the selected action For edge servers that are not the closest j , but other edge servers p hour,

[0035]

[0036] in, For edge servers j Send the task to the edge server p The time to wait for sending is the measured value after the action is run; The task is offloaded to the edge servers p The time waiting for calculation in the calculation queue is the measured value after the action is run;

[0037] (3) When the selected action For cloud servers,

[0038]

[0039] in, For edge servers j The time to wait for the task to be sent to the cloud server c is the measured value after the action is run.

[0040] S505: Use historical data to optimize the network and follow the generated actions Execute the uninstall task.

[0041] The beneficial effect of the present invention is that it provides a collaborative task offloading method in a mobile vehicle network. In order to break the misunderstanding of the traditional method of "selecting the optimal computing server from the nearby edge server", the present invention divides the task offloading process into two stages, and deploys the task offloading method on the mobile device side and the edge server side respectively. The mobile device side method decides to offload the task to the cloud server or edge computing server, and the edge server side method is responsible for rescheduling the task to find the optimal computing location, and introduces load state prediction, realizing efficient task offloading and scheduling, and realizing optimal resource configuration. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to make the purpose and technical solution of the present invention more clear, the present invention provides the following drawings for explanation:

[0043] Figure 1 This is an architecture diagram of a mobile vehicle networking system in Example 1 of the present invention; wherein 1 is a mobile device, 2 is an edge server, and 3 is a cloud server;

[0044] Figure 2 This is a flow chart of a collaborative task offloading method in a mobile vehicle network in the present invention; wherein, the thin solid arrow is the task scheduling direction, and the thick arrow is the network training feedback direction. DETAILED DESCRIPTION

[0045] In order to make the purpose and technical solution of the present invention more clearly understood, the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0046] Example 1: In order to better simulate the real driving environment, the beneficial effects of the method of the present invention are demonstrated. Figure 1 , using the Manhattan moving model to generate an area of ​​900×900m 2 The urban area is used to simulate the device movement trajectory. For the rest of the experimental settings, the consensus in the field is used, that is, data is randomly generated based on a certain probability. There are 4 mobile devices V = {v i |1≤i≤4}, 4 edge servers S={s j |1≤j≤4}, a scenario with 1 cloud server c, for any mobile device v i The parameters are Any edge server j The parameters include The parameters of cloud server c include [f c , ρ c ]; where f * Indicates the computing power of the device, that is, the number of cycles that the CPU can execute in one second; c * Represents the storage capacity of the device; ρ *Represents the total number of signal channels of the device; * indicates the corresponding device, i.e., mobile device, edge server or cloud server; and Corresponding to the mobile device v at time t i Locations and Edge Servers j location; For mobile devicesv i The speed at time t; For edge servers j The wireless network coverage radius; For edge servers j The load vector from time tn to time t, where n is a positive integer set according to the storage capacity of the device; set any task task generated by all mobile devices at time t k (t) = [e k (t), r k (t), d k (t),u k (t)], where k∈K, K is the set of tasks generated by all mobile devices, e k (t) represents the size of the task at time t, r k (t) represents the number of CPU cycles required to complete the task calculation at time t, u k (t) represents the size of the result returned after the task is calculated at time t, d k (t) represents the deadline of the task at time t, that is, at time t+d k If the result is not returned to the mobile device, the task will be considered a computation failure. An edge server is placed at each intersection in the urban area, and the coverage range of each edge server is set to 300m. The computing power of all edge servers is are uniformly sampled between (12-16), the storage capacity is uniformly sampled between (50-80), and each edge server has The cloud server has unlimited storage capacity and computing power. c is set to 30, and the number of channels ρ c 10. Place 4 mobile devices in the street and calculate the power The number of tasks k generated by the mobile device at each moment is uniformly sampled between (0-5), and the size of each task e is uniformly sampled between (2-4). k (t) is uniformly sampled between (4-8), and the required computing resources r k (t) is sampled evenly between (12-16), and the task deadline d k(t) uniformly samples between (7-15), and the result size u after completing the task calculation k (t) is 1 / 10 of the task. cell Set to 2, the transmission speed of the core network is b core Set to 7, the transmission speed of edge server communication is b v2l Set to 5.

[0047] For this scenario, in order to achieve efficient offloading of tasks generated by the mobile terminal, the present embodiment provides "a method for collaborative task offloading in a mobile vehicle network", which is implemented by a mobile vehicle network system, wherein the mobile vehicle network system comprises: 4 mobile devices, 4 edge servers, and a cloud server; the mobile device is installed on the vehicle terminal and connected to the edge server and the cloud server through a wireless network; the wireless network connecting the mobile device and the cloud server comprises two parts, a 5G network and a core network, which is first connected to the base station through the 5G network and then the base station offloads the task to the cloud server connected through the core network; the edge server is connected to the cloud server through a wired network; the mobile device, edge server, and cloud server are all computers.

[0048] Combination Figure 2 , the method comprises the following steps:

[0049] S1: The mobile device obtains the generated tasks and the length of the task waiting queue in real time.

[0050] Since the task may not be transmitted yet, incomplete data should be avoided when sampling in step S1.

[0051] S2: The mobile device calculates the time required to complete the task based on its own computing power. If the task can be completed within the deadline, the task is processed locally. Otherwise, a task offloading decision needs to be made.

[0052] The mobile device v described in step S2 at time t i Calculate its task based on its own computing power k (t) Time required for completion Among them, mobile devices v i Computing tasks k Waiting time (t) x i，j / c,k Represents a mobile device v i Uninstall task task k To edge servers j Or cloud server c; if the device decides to offload the task to an edge server or cloud server, then x i,j / c,k =1; on the contrary, when deciding the local computing task, xi，j / c,k = 0. If the mobile device has enough storage space and T c ≤d k On mobile devices v i The task is processed locally.

[0053] S3: If the mobile device is outside the coverage area of ​​the edge server, or the task cannot be offloaded before the mobile device leaves the coverage area of ​​the edge server, the task is directly offloaded to the cloud server.

[0054] Specifically: the distance between the mobile device and all edge servers is calculated according to the Euclidean distance formula. If these distances are greater than the wireless network coverage radius of all edge servers, the task is directly unloaded to the cloud server. At the same time, if the wireless network coverage requirements are met, the location at the task deadline is calculated based on the speed of the mobile device and the task deadline. If the location is out of the wireless network coverage of all edge servers, the task is directly unloaded to the cloud server.

[0055] S4: When task offloading decision is needed, the location information of the mobile device and the edge server and the time information of sending the task are combined to use the DDQN-based mobile device-side task offloading method to decide whether to offload the task to the edge server or the cloud server.

[0056] The DDQN-based mobile device side task offloading method is implemented using the DDQN network. i , specifically:

[0057] S401: Calculate the distance between the mobile device and v according to the Euclidean distance formula i Nearest Edge Servers j ;

[0058] S402: Calculate the maximum transmission rate b of a single channel using the Shannon formula constraint, and further calculate the edge server s j Transmission speed

[0059] 5403: Combine the location information of the mobile device and the edge server and the time information of sending the task to obtain the status of the task offloading on the mobile device side in; For edge servers j Computing tasks k The waiting time (t) can be calculated based on the edge server s j The sum of the computation time of the tasks in the queue is calculated;

[0060] S404: Set rewards

[0061] S405: Combine the status of task offloading on the mobile device side and rewards Selecting actions using the DDQN network

[0062] S406: Follow the action Perform task offloading.

[0063] S5: When the task is offloaded to the edge server, the A3C-based distributed method will determine the actual execution location of the task according to the overall load status of the server cluster.

[0064] The A3C-based distributed method described in step S5 is implemented using the A3C network, specifically:

[0065] S501: Predict the load vector of all edge servers at the next moment based on the historical load of all edge servers

[0066] Any edge server j Based on the load from time tn to time t, the predicted load at time t+1 is calculated as Among them, f θ (·) is a neural network with parameter θ;

[0067] S502: Obtaining the status of the A3C network Where T is the time when the task is generated;

[0068] S503: Selecting an action using the Actor network of the A3C network in, are positive integers, corresponding to the codes of the edge server and the cloud server respectively;

[0069] S504: Calculate reward based on task completion time setting

[0070] Specifically, (1) when the selected action is the distance v from the mobile device i Nearest Edge Servers j hour,

[0071]

[0072] Among them, φ k Indicates whether the mobile device is within the coverage of the current server after completing the task calculation. If yes, it is 0, otherwise it is 1; Indicates that the task is offloaded to the edge server s j The time waiting for calculation in the calculation queue is the measured value after the action is run;

[0073] (2) When the selected action For edge servers that are not the closest j , but other edge servers p hour,

[0074]

[0075] in, For edge servers j Send the task to the edge server p The time to wait for sending is the measured value after the action is run; The task is offloaded to the edge servers p The time waiting for calculation in the calculation queue is the measured value after the action is run;

[0076] (3) When the selected action For cloud servers,

[0077]

[0078] in, For edge servers j The time to wait for sending the task to the cloud server c is the measured value after the action is run;

[0079] S505: Use historical data to optimize the network and follow the generated actions Execute the uninstall task.

[0080] S6: Execute the computing task and send the result back to the mobile device.

[0081] In order to better demonstrate the beneficial effects of the method of the present invention, this embodiment conducts a comparative experiment with the prior art DRL [1]. The DRL method combines the DDQN and AC algorithms. When a task is generated, the mobile device directly decides the execution location of the task. Because this method assumes that the CPU is used by multiple tasks at the same time, it is necessary to know the number of currently active tasks, so an active queue number prediction algorithm is added.

[0082] [1] TANG M, WONG VW S. Deep reinforcement learning for task offloading in mobile edge computing systems [J]. IEEE Transactions on Mobile Computing, 2020, 21(6): 1985-1997.

[0083] The evaluation indicators used in this field are task completion rate and weighted average task completion time to measure the performance of the algorithm. Where N′ is the number of completed tasks, N is the total number of tasks; weighted average task completion time WACT = TCR·ACT+(1-TCR)·d avg , where ACT is the average completion time of N′ tasks, and d avg is the average of the deadlines of N tasks. Since DRL models the entire environment, its overall performance is close to that of the method of the present invention, but there is still a gap. In more detail, according to the experimental results, the DRL algorithm offloads approximately 38.42% of the tasks to the edge server, and only 23% of the tasks are sent to the cloud server, causing the edge server to be overloaded, while the cloud server with more powerful computing power does not play a greater role. In contrast, the method of the present invention offloads 24.4% of the tasks to the edge server, and approximately 44.17% of the tasks are offloaded to the cloud server. Compared with DRL, the task completion rate of the method of the present invention is increased by 8%, and the task completion time is reduced by 13.2%. .

[0084] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A mobile vehicle networking system, characterized in that: The mobile vehicle networking system includes: several mobile devices, several edge servers, and a cloud server; the mobile device is installed on the car terminal and connected to the edge server and the cloud server through a wireless network; the wireless network connecting the mobile device and the cloud server includes two parts: a 5G network and a core network, which is first connected to the base station through the 5G network and then the base station offloads the task to the cloud server connected through the core network; the edge server is connected to the cloud server through a wired network; the mobile device, edge server, and cloud server are all computers.

2. A collaborative task offloading method in a mobile vehicle network as claimed in claim 1, characterized in that: For M mobile devices V = {v i |1≤i≤M}, N edge servers S={s j |1≤j≤N}, a scenario with 1 cloud server c, for any mobile device v i The parameters are Any edge server j The parameters include The parameters of cloud server c include [f c , ρ c ]; where f * Indicates the computing power of the device, that is, the number of cycles that the CPU can execute in one second; c * Represents the storage capacity of the device; ρ * Represents the total number of signal channels of the device; * indicates the corresponding device, i.e., mobile device, edge server or cloud server; and Corresponding to the mobile device v at time t i Locations and Edge Servers j location; For mobile devicesv i The speed at time t; For edge servers j The wireless network coverage radius; For edge servers j The load vector from time tn to time t, where n is a positive integer set according to the storage capacity of the device; set any task task generated by all mobile devices at time t k (t) = [e k (t), r k (t), d k (t),u k (t)], where k∈K, K is the set of tasks generated by all mobile devices, e k (t) represents the size of the task at time t, r k (t) represents the number of CPU cycles required to complete the task calculation at time t, u k (t) represents the size of the result returned after the task is calculated at time t, d k (t) represents the deadline of the task at time t; the 5G network transmission speed is recorded as b cell , the transmission speed of the core network is recorded as b core ; The method comprises the following steps: S1: The mobile device obtains the generated tasks and the length of the task waiting queue in real time; S2: The mobile device calculates the time required to complete the task based on its own computing power. If the task can be completed within the deadline, the task is processed locally. Otherwise, a task offloading decision needs to be made. S3: If the mobile device is outside the coverage area of ​​the edge server, or the task cannot be offloaded before the mobile device leaves the coverage area of ​​the edge server, the task is directly offloaded to the cloud server; S4: When task offloading decision is needed, the mobile device-side task offloading method based on DDQN is used to decide whether to offload the task to the edge server or the cloud server, combining the location information of the mobile device and the edge server and the time information of sending the task; S5: When the task is offloaded to the edge server, the distributed method based on A3C (Asynchronous Advantage Actor-Critic, A3C for short) will determine the actual execution location of the task according to the overall load status of the server cluster; S6: Execute the computing task and send the result back to the mobile device; Further, the DDQN-based mobile device side task offloading method described in step S4 is implemented using the DDQN network, for the mobile device v at time t i , specifically: S401: Calculate the distance between the mobile device and v according to the Euclidean distance formula i Nearest Edge Servers j ; S402: Calculate the maximum transmission rate b of a single channel using the Shannon formula constraint, and further calculate the edge server s j Transmission speed S403: Acquire the status of task offloading on the mobile device side by combining the location information of the mobile device and the edge server and the task sending time information in; For edge servers j Computing tasks k The waiting time (t) can be calculated based on the edge server s j The sum of the computation time of the tasks in the queue is calculated; S404: Set rewards S405: Combine the status of task offloading on the mobile device side and rewards Using DDQN (Double Q-learning Networks, DDQN for short) to select actions S406: Follow the action Perform task offloading; Further, the A3C-based distributed method described in step S5 is implemented using the A3C network, specifically: S501: Predict the load vector of all edge servers at the next moment based on the historical load of all edge servers S502: Obtaining the status of the A3C network Where T is the time when the task is generated; S503: Selecting an action using the Actor network of the A3C network in, are positive integers, corresponding to the codes of the edge server and the cloud server respectively; S504: Calculate reward based on task completion time setting S505: Use historical data to optimize the network and follow the generated actions Execute the uninstall task.

3. The method for unloading collaborative tasks in a mobile vehicle network according to claim 2, characterized in that: The mobile device v described in step S2 at time t i Calculate its task based on its own computing power k (t) Time required for completion Among them, mobile devices v i Computing tasks k Waiting time (t) x i,j / c,k Represents a mobile device v i Uninstall task task k To edge servers j Or cloud server c; if the device decides to offload the task to an edge server or cloud server, then x i，j / c,k =1; on the contrary, when deciding the local computing task, x i,j / c,k = 0; if the mobile device has enough storage space and T c ≤d k On mobile devices v i The task is processed locally.

4. The method for unloading collaborative tasks in a mobile vehicle network according to claim 2, characterized in that: Step S3 is specifically as follows: the distance between the mobile device and all edge servers is calculated according to the Euclidean distance formula. If these distances are greater than the wireless network coverage radius of all edge servers, the task is directly unloaded to the cloud server; at the same time, if the wireless network coverage requirements are met, the location at the task deadline is calculated according to the speed of the mobile device and the task deadline. If the location is out of the wireless network coverage of all edge servers, the task is directly unloaded to the cloud server.

5. The method for unloading collaborative tasks in a mobile vehicle network according to claim 2, characterized in that: In step S501, any edge server s j The predicted load at time t+1 is calculated based on the load from time tn to time t: Among them, f θ (·) is a neural network with parameter θ.

6. A method for unloading collaborative tasks in a mobile vehicle network according to claim 2, characterized in that: The step S504 specifically includes the following steps: (1) When the selected action is the distance v from the mobile device i Nearest Edge Servers j hour, Among them, φ k Indicates whether the mobile device is within the coverage of the current server after completing the task calculation. If yes, it is 0, otherwise it is 1; Indicates that the task is offloaded to the edge server s j The time waiting for calculation in the calculation queue is the measured value after the action is run; (2) When the selected action For edge servers that are not the closest j , but other edge servers p hour, in, For edge servers j Send the task to the edge server p The time to wait for sending is the measured value after the action is run; The task is offloaded to the edge servers p The time waiting for calculation in the calculation queue is the measured value after the action is run; (3) When the selected action For cloud servers, in, For edge servers j The time to wait for the task to be sent to the cloud server c is the measured value after the action is run.

Images